Search Results (235)

Node-level cargo-volume forecasting in logistics sorting networks requires modeling temporal dynamics together with directed inter-node dependencies and planned topology perturbations. This study addresses 1-h-ahead forecasting under a partial-information boundary, where historical node volumes, the pre-change network structure, and planned route-topology changes are available before prediction, whereas continuous post-change dynamic edge weights and realized post-change graph states are unavailable. We propose a perturbation-aware framework that represents the sorting system as a directed network and integrates temporal features, pre-change structural descriptors, topology-change encodings, perturbation-response proxies, and similarity-assisted support for data-scarce nodes within a unified forecasting protocol. A shared random forest backbone is used only to assess the incremental value of these representations. Experiments on 57 sorting centers show that temporal dynamics dominate under stable-network conditions. Under topology perturbation, topology-change signals reduce test weighted absolute percentage error (WAPE) from 18.10% to 17.11%, and perturbation-response proxies further reduce it to 16.91%. For data-scarce nodes, similarity support reduces test WAPE from 29.43% to 26.68%, with consistent gains under 10%, 20%, and 30% sample-retention settings. These results suggest that the framework provides an interpretable and information-admissible representation strategy for node-level forecasting in directed networked systems. Full article

Thehitting time index $\mathrm{HT}\left(G\right)$ is a recently introduced topological descriptor based on expected hitting times of a random walk on a graph. In this paper, we derive a closed-form formula for $\mathrm{HT}\left(G\right)$ for broom graphs $B_{n,d}$ that holds for all parameters $2\le d\le n-1$, $\mathrm{HT}\left(B_{n,d}\right)=S_1+S_2+S_3+(n-d){\displaystyle \sum _{i=1}^{d-1}}\max \{A\left(i\right),B\left(i\right)\},$ where $S_1,S_2,S_3,A\left(i\right),B\left(i\right)$ are explicitly defined. For $d\ge 2$ and $n\ge 4d-8$ we derive a simpler cubic polynomial formula in n, $\mathrm{HT}\left(B_{n,d}\right)=n^3+a_d{n}^2+b_{d}n+c_d,$ with explicitly given coefficients $a_d,b_d,c_d$ depending only on d. We also consider quartic polynomial formulas for special cases. Full article

Background/Objectives: Coagulation factor XIa (FXIa) has emerged as a promising therapeutic target for the development of safer anticoagulant therapies with reduced bleeding risk. This study aimed to develop an interpretable machine learning-driven quantitative structure–activity relationship (QSAR) framework for predicting the inhibitory activity of FXIa inhibitors and supporting virtual screening applications. Methods: A total of 3026 curated compounds retrieved from the ChEMBL database were used for regression modeling, whereas 2119 compounds were retained for classification modeling after excluding intermediate-activity molecules. Molecular descriptors were generated using RDKit, Mordred, and Morgan fingerprint representations. Following preprocessing and feature selection, multiple machine learning algorithms were systematically benchmarked. Model robustness and reliability were further evaluated using 5-fold cross-validation, scaffold-aware validation, applicability domain analysis, and Y-randomization testing. Results: Nonlinear ensemble learning approaches consistently outperformed conventional linear algorithms. The optimized HistGradientBoostingRegressor achieved the best regression performance, with an independent test-set R² value of 0.711 and an RMSE value of 0.759, whereas the optimized classification model achieved accuracies approaching 95%. SHAP analysis identified lipophilicity-related descriptors, aromatic scaffold organization, electrostatic surface properties, and molecular topology as major contributors to FXIa inhibitory activity prediction. In addition, a proof-of-concept virtual screening workflow successfully identified several candidate compounds exhibiting high predicted pKi values and elevated active-class probabilities. Conclusions: The proposed framework provides a robust, interpretable, and reproducible machine learning-driven QSAR strategy for FXIa inhibitor discovery and may facilitate future virtual screening campaigns and medicinal chemistry optimization studies targeting FXIa-associated anticoagulant drug discovery. Full article

Zeolites, both natural (e.g., clinoptilolite) and synthetic (e.g., FAU, ZSM-5), provide robust, tunable platforms for the removal of air pollutants and process-stream contaminants via adsorption and catalysis. This author-led article integrates experimental and theoretical insights on the adsorption of odorous compounds and ammonia (NH₃) and the catalytic abatement of nitrogen oxides (NOx) and nitrous oxide (N₂O), highlighting how topology, acidity, and metal speciation jointly control performance. Representative theoretical results show that adsorption on Brønsted acid sites is significantly more favorable (≈−1.1 eV for NH₃ and −0.37 eV for acetaldehyde) than on Na⁺ sites (≈0.02 eV and 1.22 eV, respectively), demonstrating the critical role of acid site distribution in adsorption selectivity. We dissect structure–function relationships encompassing pore size and connectivity, Si/Al ratio, Brønsted/Lewis site distribution, hydrophilicity/hydrophobicity, and the role of water, with emphasis on hierarchical porosity to alleviate transport limitations. Metal exchange and surface functionalization are discussed as levers to tailor adsorption strength and redox activity, supported by density functional theory (DFT) analyses and reaction pathways. We propose practical design descriptors (acid strength metrics, metal nuclearity, and confinement factors) that enable faster iteration of zeolite architecture for targeted separations and reactions. Sustainability considerations include the use of abundant natural zeolites, low-energy regeneration, stability under humid, mixed-stream conditions that minimize pressure drop and waste. The article closes with a forward look at data-guided optimization to accelerate “engineering zeolites” for durable, selective, and energy-efficient clean-air and process-intensification applications. Full article

Road network extraction and updating are crucial for urban development, map updating, and mobility applications. Existing trajectory-based methods often underutilize grid-level semantic information and neighborhood context, thereby limiting their robustness to noisy, heterogeneous, and cross-city trajectory conditions. This study proposes a supervised framework for trajectory-driven road network extraction by coupling intra-grid movement semantics with inter-grid neighborhood context. Multi-level features, including convex-hull shape descriptors, directional clustering, DTW-based (Dynamic Time Warping) heterogeneity, and neighborhood density differences, are used to train a Random Forest classifier for key-grid detection. The detected key grids are further processed through morphology-aware thinning and Kalman smoothing to generate a topology-preserving and vectorization-ready road skeleton. The model is trained on pedestrian trajectories from Shenzhen and directly transferred to vehicle trajectories in Wuhan and Changsha under a zero-shot setting. Experimental results show that the proposed method achieves longer correctly extracted road length and competitive length-based precision compared with raster-based reference methods, while feature-importance and ablation analyses confirm the complementary role of neighborhood context. The proposed pipeline is scalable, interpretable, and transferable, supporting trajectory-based road map updating and urban network analysis. Full article

Symmetry plays a fundamental role in the structural analysis of lattice-based systems, particularly in graphene-like molecular structures. In chemical graph theory, counting independent sets is equivalent to computing the Merrifield–Simmons (M–S) index, a key descriptor of molecular stability in conjugated systems. Most existing exact counting methods rely on regular lattice symmetry, where structural uniformity simplifies computation; however, these approaches are difficult to extend to irregular graphs, where symmetry breaking introduces non-local dependencies and increases computational complexity. This paper proposes an asymmetry-aware algorithmic framework based on Hamiltonian traversals and a traversal-induced pathwidth parameter $w\left(G\right)$, defined through backward dependencies. Our method organizes non-local adjacencies into a bounded set of structured constraints, enabling a dynamic programming scheme over a reduced state space. The resulting algorithm runs in time $\mathcal{O}\left(2^{w\left(G\right)}·\mathrm{poly}\left(n\right)\right)$ and is fixed-parameter tractable with respect to $w\left(G\right)$. The results demonstrate that asymmetry-aware traversal strategies enable efficient exact enumeration in irregular mesh graph families, providing a robust computational framework for analyzing molecular descriptors in graphene-based structures with topological defects such as Stone–Wales transformations. Full article

Molecular property prediction is a fundamental task in drug discovery and materials design. While graph neural networks (GNNs) and SMILES-based Transformers have made significant strides, the former are often limited by local message-passing bottlenecks such as over-squashing, while the latter frequently lack explicit topological constraints and suffer from severe vocabulary imbalance. In this work, we revisit the granularity of molecular modeling and propose a representation learning framework built upon bond-level sequences. Our framework models molecules as sequences of directed bond tokens and introduces a structure-aware hybrid attention mechanism. By imposing hard topological constraints on a subset of attention heads to reinforce local connectivity while preserving global receptive fields in the remaining heads, the design is intended to separate short-range chemical bonding from long-range contextual dependencies. For pre-training, we implemented a multi-scale consistency learning paradigm, which utilizes an atom-centric group masking strategy to induce a hierarchical loss of local structural information and employs contrastive and triplet losses to ensure identity consistency across varying scales of structural degradation. Furthermore, by incorporating macro-scale physicochemical descriptors (e.g., LogP, TPSA) as global anchors, we examined how the inclusion of global attribute bias can provide weak physicochemical priors during pre-training, while its effect during downstream fine-tuning remains task-dependent. Experimental results demonstrate that our lightweight model, with approximately 3.5 million parameters, exhibits a dataset-dependent performance profile across MoleculeNet benchmarks and shows promising behavior on selected topology-sensitive tasks, particularly MUV. Ablation studies further analyze the contribution of bond-level connectivity, the stage-dependent dynamics of global attribute bias, structured masking, and pre-training configurations. Ultimately, this work provides an alternative representation design for molecular modeling, offering a parameter-efficient option for future molecular learning systems alongside traditional SMILES-based and graph-based formulations. Full article

We propose a topological data analysis framework for the study of damage evolution in anisotropic composite materials based on scalar filtrations defined on cubical complexes. Two complementary anisotropic filtrations are constructed from the structure tensor: a fibre-oriented filtration f1, capturing directional coherence, and a crack-oriented filtration f2, sensitive to isotropic and weakly oriented structures. Zero-dimensional persistent homology is analysed through merge trees built from the superlevel-set filtration via the transformation $g=1-f$, providing a hierarchical representation of connected components. Higher-order connectivity is described using skeleton-based Reeb-like graphs. From these constructions, we derive spatial and global descriptors, including a topological danger map and a Topological Damage Complexity Index (TDCI) based on one-dimensional persistent homology. The behaviour of the TDCI is examined with respect to variations in its parameters and to image perturbations, showing consistent trends across the considered configurations. The results highlight complementary structural behaviours captured by the two filtrations and show a coherent correspondence with observed patterns. Overall, the proposed framework provides a mathematically grounded description of structural organisation. It is intended as an exploratory approach, and further work is needed to clarify its relationship with the underlying physical damage mechanisms. Full article

Nonlinear dynamical systems exhibit complex behaviors such as periodicity and chaos, which are traditionally analyzed using time-series data. However, these approaches often fail to capture the intrinsic geometric structure of the system dynamics represented in the phase space. In this study, we address this limitation by proposing a topological machine learning framework that leverages phase portrait images to classify dynamical regimes. The primary objective of this study is to investigate whether the topological features extracted from phase portraits can effectively distinguish between periodic and chaotic behaviors across different nonlinear systems. To achieve this, we employed the Topological Data Analysis (TDA) technique of cubical homology, which enables the extraction of topological descriptors, such as persistence diagrams and Betti curves. We used these features to train multiple machine learning (ML) classifiers, including XGBoost, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Gaussian Naïve Bayes (GNB), and Random Forest (RF). The experimental results across benchmark systems, including the Chua, Lorenz, Mathieu–Duffing, and erbium-doped fiber laser models, demonstrate that the proposed approach achieves high classification accuracy, with performance improving from approximately 93% under $H_0$ features to 99–100% under $H_1$ and combined feature representations. These findings highlight that topological features, particularly $H_1$, effectively capture the underlying geometric structure of dynamical systems. Overall, the proposed framework provides a robust, interpretable, and generalizable approach for phase portrait classification, with potential applications in nonlinear system analysis, pattern recognition, and early detection of chaotic transitions. Full article

Immune checkpoint modulation has emerged as a promising strategy in cancer therapy, including the treatment of aggressive tumors such as glioblastoma. Among these targets, programmed death-ligand 1 (PD-L1) plays a key role in tumor immune evasion and represents an attractive target for small-molecule inhibitor development. In this study, a virtual screening approach was applied to identify potential PD-L1 modulators within a library of nucleoside-related compounds and structurally similar molecules. A dataset of 400 compounds was evaluated using molecular docking to predict their binding affinity (free energy values and binding pose) toward PD-L1. The resulting complexes were analyzed to identify nonbond interactions within the hydrophobic pocket formed at the PD-L1 dimer interface. In addition to docking results, physicochemical descriptors associated with drug-likeness and blood-brain barrier penetration were calculated, including lipophilicity, molecular weight, hydrogen bond donors and acceptors, as well as topological polar surface area. To integrate these parameters, a multiparameter optimization (MPO) score was implemented. Finally, molecular dynamics simulations of protein-ligand interactions were performed to explore the structural stability for 100 ns using the most promising ligands. The analysis revealed that several top-ranked compounds exhibited favorable docking scores and physicochemical properties compatible with drug-like behavior. Interestingly, BMS-1, a known PD-L1 inhibitor, was identified among the highest-scoring compounds, supporting the reliability of the MPO protocol. Furthermore, multiple candidates displaying nucleoside-like scaffolds combined with reduced polarity and moderate lipophilicity emerged as promising molecules according to the MPO ranking. Overall, the results suggest that nucleoside-derived scaffolds may represent a viable starting point for the development of small-molecule PD-L1 modulators with potential applicability in glioblastoma therapy. Full article

Accurate anatomical landmark detection is important for orthodontic analysis, surgical planning, and morphometric measurement, but fully supervised methods usually require large expert-annotated datasets. This work studies a one-shot setting, where only a single annotated template image is used for training. We propose a foundation-model-based landmark detection framework using a frozen DINO Vision Transformer (ViT) backbone. The proposed framework integrates three complementary components: a Multi-Layer Multi-Facet (MLMF) module that adaptively fuses key and value features from multiple ViT layers through global source-wise reweighting; a Mamba-Based Long-Range Context Aggregation (MLCA) module that injects global anatomical context into fused patch descriptors with linear complexity; and a Topology-Constrained Graph Refinement (TCGR) module that refines the predicted landmark configuration using anatomical graph constraints. Experiments on the Cephalometric dataset and the Hand X-ray dataset demonstrate that the proposed method achieves strong performance. Overall, the results show that jointly exploiting multi-source foundation-model representations, efficient long-range context aggregation, and topology-aware refinement improves annotation-efficient anatomical landmark detection. Full article

Quantitative structure–activity relationship (QSAR) modeling has conventionally relied on expert-designed molecular descriptors to encode chemical structures. DeepSnap is a descriptor-free QSAR approach that converts prepared three-dimensional molecular conformers into image representations and feeds them directly into convolutional neural networks for activity prediction. This focused narrative review traces DeepSnap from its introduction in 2018 to its current state and places it within the broader landscape of descriptor-based QSAR, topology-based and 3D-aware graph neural networks, and related image-based or semi-image-based molecular representation approaches. Previous studies applied DeepSnap to Tox21 nuclear receptor and molecular initiating event endpoints, rat hepatic clearance, blood–brain barrier penetration, acute oral toxicity, and cosmetics–pharmaceutical compound classification. Across the DeepSnap series, image-based and descriptor-based predictions have provided complementary information, particularly in ensemble or consensus models. However, high or near-ceiling ROC–AUC values reported for selected endpoints should not be interpreted as indicating deterministic or universally generalizable predictions; rather, they should be considered in the context of endpoint-specific model development, image-rendering parameter optimization, possible class imbalance, split dependence, limited matched external replication, and incomplete benchmarking against modern molecular representation models. Limitations include a dependence on nonphysical rendering parameters, single- or representative-conformer input, incomplete matched benchmarking against 2D and 3D molecular representation models, and an interpretability gap addressed in part by CAM-family visualization in the AI-based Substance Hazard Integrated Prediction System (AI-SHIPS) and S-COPHY (a model developed by Shiseido for cosmetics–pharmaceutical compound classification). Future directions include standardized image-generation protocols, conformer-ensemble extensions, systematic interpretability analysis, matched benchmarking, and potential integration with graph-based and 3D-aware molecular learning approaches. Full article

Background: The blood–brain barrier (BBB), which restricts the brain penetration of most small molecules and almost all biologics, continues to be a significant hurdle in the development of drugs for the central nervous system (CNS). During early-stage screening, a reliable computational prediction of BBB permeability, typically expressed as log BB, can help reduce the experimental load. Methods: We provide a well-validated machine learning system created solely using the B3DB experimental database, which includes 7807 chemicals with BBB⁺/BBB⁻ annotations and 1058 compounds with in vivo log BB values. Using the Mordred library, a carefully selected set of 40 two-dimensional chemical descriptors was calculated from SMILES notation without the use of artificial data augmentation. Stratified five-fold cross-validation was used to comprehensively benchmark the nine methods used in this study. Results: On a held-out test set (n = 212), gradient boosting produced the greatest regression performance, with R² = 0.6043, RMSE = 0.4740 log units, and MAE = 0.3326, which is in line with the upper range recorded for experimental BBB datasets. On an internal test set (n = 1562), the corresponding classifier obtained an AUC-ROC of 0.9476 and a balanced accuracy of 0.8568; on an independent external validation set (n = 175), it achieved an AUC-ROC of 0.9137. Topological polar surface area was found by SHAP analysis to be the primary factor influencing BBB permeability, with lipophilicity and ionization-related characteristics being the second and third most important factors, respectively. Nonlinear relationships in accordance with accepted pharmacokinetic principles were validated using partial dependence analysis. Conclusion: This study provides a reliable technique for predicting BBB permeability in CNS drug discovery. Full article

In the present study, we built several models based on three-dimensional molecular descriptors to predict the glass transition temperature using a data set of 117 polymers. A data set division was established (training and test data) and consequently the models were developed and validated. Finally, the prediction/screen of the top models were compared. Three main descriptors were obtained with excellent predictions: E2 (E2u and E2s), which encodes angular and radial information about atomic configuration, usually in relation to two atoms; TDB (TDB10u, TDB10e, TDB10s) describes the relationship between the average three-dimensional (Euclidean) distance and the topological distance (path length, or number of bonds) between possible atom pairs in a molecule; and RDF (RDF25i, RDF65u, RDF25u) describes the density of atoms at different distances from a reference atom, capturing information about the local structure of the molecule. An initial exploratory GA-LDA classification analysis using 3D descriptors revealed only partial discrimination between polymers with distinct T_g behavior, indicating that simplified 3D structural representations alone are limited for robust T_g prediction. Consequently, graph-based (2D) descriptors models were created and the prediction of the T_g was successfully achieved. Overall, the most influential variables are predominantly graph-based (2D) descriptors associated with molecular connectivity patterns (e.g., autocorrelation-type descriptors such as ATS2*), topological/shape-related indices (TSC* family), and ring-related terms. This attribution profile is consistent with the expected physicochemical determinants of the glass transition: polymer repeat units with higher structural rigidity, constrained connectivity, and increased ring/unsaturation content that typically exhibits reduced segmental mobility and, therefore, higher T_g. Full article

The rapid advancement of Machine Learning (ML) has significantly transformed polymer science by enabling efficient prediction and design of polymer properties through high-throughput screening. However, current methods still struggle with nonlinear Structure–Property Relationships (SPRs), limited dataset standardization, and computational inefficiency, which restrict prediction accuracy and interpretability. This study proposes a comprehensive ML-based framework for predicting polymer properties and identifying SPRs. The approach integrates data preprocessing, molecular descriptor and topological index–based feature extraction, iterative feature selection, and XGBoost predictive modeling. Model hyperparameters are optimized using the Starfish Optimization Algorithm (SOA) to enhance performance and efficiency. Model interpretability is achieved through SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), providing both global and local insights into the influence of molecular features on polymer properties. Experimental evaluation on the PolyOne dataset demonstrates strong predictive performance, with R² values exceeding 0.92, mean absolute error (MAE) below 0.08, and root mean square error (RMSE) under 0.12 for key physical and optical polymer properties. Overall, the proposed framework effectively balances accuracy, computational efficiency, and interpretability, offering a robust and practical tool for accelerating polymer design while enhancing understanding of molecular structure–property relationships. Full article